Parking lot revenue increase opportunity identification and package generation method and device, and storage medium

Abstract

The application discloses a parking lot income-increasing opportunity identification and package generation method and device and a storage medium, and comprises the following steps: acquiring multi-source operation data of a target parking lot and extracting multi-dimensional features from the multi-source operation data; determining the income-increasing opportunity level of the target parking lot based on the multi-dimensional features; predicting the idle time period distribution and releasable capacity in a future preset period according to time sequence related features in the multi-dimensional features; generating a parking package draft and pre-filling package parameters according to the income-increasing opportunity level, the idle time period distribution and the releasable capacity; performing verification on the parking package draft, wherein the verification at least includes time period conflict verification and capacity constraint verification; and outputting the parking package draft after the verification is passed. Through multi-dimensional feature fusion identification and idle time period time sequence prediction, the application realizes automatic generation of a package draft and parameter verification, so that manual judgment is replaced by automatic data driving, and the response speed of the income-increasing opportunity is improved.

Description

Technical Field

[0001] This application relates to the field of intelligent parking algorithm technology, and in particular to a method, device and storage medium for identifying parking lot revenue-generating opportunities and generating packages. Background Technology

[0002] With the continuous growth of urban motor vehicle ownership and the near saturation of parking resource supply, the parking industry has shifted from a phase of incremental construction to a phase of stock management. Parking space utilization rates vary significantly across different time periods, including daytime, nighttime, weekdays, weekends, and holidays. A large amount of idle resources have not been effectively commercialized, resulting in the underutilization of revenue growth potential. How to systematically tap into the scattered idle resources of various parking lots and transform them into tradable commodities has become a critical issue that urgently needs to be addressed in the parking management sector.

[0003] In terms of parking operation and management technology, known solutions include those that acquire parking statistics within a preset time period to identify parking behavior patterns of different user groups and generate corresponding parking packages. Other solutions compare and analyze current and historical operational data to identify discrepancies in operational indicators, thereby pinpointing abnormal indicators and generating corresponding operational strategy recommendations. Regarding revenue analysis, some solutions identify factors influencing parking revenue and categorize them according to their direction and degree of impact, pinpointing the root cause when revenue fluctuations exceed set thresholds.

[0004] However, the above solutions still have significant limitations in practical applications. Identifying revenue-generating opportunities typically relies on manual judgment by operations staff based on entry and exit reports, vacancy rate statistics, and market experience. This makes it difficult to promptly and accurately identify scattered revenue-generating opportunities across a large number of parking lots. Furthermore, the analysis of idle periods lacks the ability to accurately predict time distribution and available capacity, making it difficult to dynamically adjust package plans according to regional supply and demand changes. This results in a lag in the response speed to revenue-generating opportunities compared to market changes.

[0005] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0006] The main purpose of this application is to provide a method, device and storage medium for identifying parking revenue-generating opportunities and generating packages, which aims to solve the technical problem that the identification of revenue-generating opportunities and the generation of packages in the existing parking resource revenue-generating management process rely heavily on manual judgment and lack an automated closed-loop processing mechanism.

[0007] To achieve the above objectives, this application proposes a method for identifying parking lot revenue-generating opportunities and generating service packages, the method comprising: Acquire multi-source operational data of the target parking lot, and extract multi-dimensional features from the multi-source operational data; The revenue-generating opportunity level of the target parking lot is determined based on the aforementioned multi-dimensional features; Based on the time-related features in the multi-dimensional features, predict the distribution of idle time periods and releaseable capacity within a future preset period; Based on the revenue-increasing opportunity level, the idle time distribution, and the available capacity, a draft parking package is generated and package parameters are pre-filled. The proposed parking package is validated, and the validation includes at least time period conflict validation and capacity constraint validation. Output the draft parking package after verification.

[0008] In one embodiment, the step of determining the revenue-generating opportunity level of the target parking lot based on the multi-dimensional features includes: Normalization is performed on each dimension of the multi-dimensional features to obtain the normalized score of each dimension. The multi-dimensional features include time period vacancy rate feature, time period repetition feature, regional demand heat feature, price sensitivity feature and income fluctuation feature. Each dimension is assigned a weight, which is determined by at least one of the following methods: expert weighting, entropy weighting, or data-driven weighting. The comprehensive score is calculated by weighting and summing the normalized scores of each dimension with their corresponding weights. Based on the comprehensive score and the preset classification threshold, the target parking lot is divided into multiple revenue-increasing opportunity levels, including high opportunity, medium opportunity and low opportunity. The corresponding listing priority is determined based on the revenue increase opportunity level obtained from the classification. The high opportunity corresponds to the high listing priority, the medium opportunity corresponds to the medium listing priority, and the low opportunity corresponds to the low listing priority or postponement of listing.

[0009] In one embodiment, after the step of dividing the target parking lot into multiple revenue-generating opportunity levels based on the comprehensive score and a preset classification threshold, the method further includes: Obtain the normalized score of at least one key dimension of the target parking lot; When the normalized score of the key dimension is lower than the preset lower limit, the income increase opportunity level will be downgraded by one or more levels. The key dimension includes at least the regional demand heat feature.

[0010] In one embodiment, the step of predicting the distribution of idle time periods and releaseable capacity within a future preset period based on the time-series correlation features among the multi-dimensional features includes: Obtain historical parking space occupancy rate time series data of the target parking lot; Based on the historical parking space occupancy rate time series data and calendar features, the idle probability or number of idle parking spaces at each time granularity within a future preset period is predicted using a time series prediction model. Consecutive or adjacent idle time slices are aggregated into idle time intervals, each of which includes a start and end time. The corresponding release capacity is determined based on the number of idle parking spaces within each idle time period interval, and the calendar features include at least one of time, day of the week, weekday identifier, weekend identifier, and holiday identifier.

[0011] In one embodiment, the step of predicting the probability of vacancy or the number of vacant parking spaces at each time granularity within a future preset period using a time-series prediction model based on the historical parking space occupancy rate time-series data and calendar features includes: Obtain regional demand data, surrounding points of interest data, and vacancy data of competitor parking lots for the target parking lot; The regional demand data, the surrounding points of interest data, and the vacancy data of competitor parking lots are used as exogenous variables and input into the time series prediction model. The time series prediction model is any one of a time series statistical model, a gradient boosting model, or a recurrent neural network. When the parking space occupancy data of the target parking lot shows periodicity and stable regularity, the time series statistical model is selected. When the input feature dimension contains exogenous variables and there is a non-linear interaction relationship between features, the gradient boosting model is selected. When the time series has a long-range dependency relationship and the sample size is sufficient, the recurrent neural network is selected.

[0012] In one embodiment, the step of generating a draft parking package and pre-filling package parameters based on the revenue-increasing opportunity level, the idle time distribution, and the available capacity includes: The package type and package priority are determined based on the aforementioned revenue-increasing opportunity level; The idle time periods are categorized according to preset time period types and mapped to at least one of weekday daytime periods, weekday nighttime periods, weekend periods, or holiday periods. Continuous or adjacent idle time periods are aggregated into time periods applicable to the package. The saleable capacity of the package is calculated based on the releasable capacity. The saleable capacity of the package is obtained by multiplying the releasable capacity by a reduction factor and deducting the safety buffer, then rounding down. A benchmark price is determined based on the hourly rental price of the target parking lot's main business and the prices of similar parking resources in the surrounding market. A price adjustment coefficient is calculated based on the regional demand characteristics, the time period vacancy rate characteristics, and the price sensitivity characteristics. A package price range is generated based on the benchmark price and the price adjustment coefficient. The applicable time period, available capacity, price range, and performance restriction parameters of the parking package draft are pre-filled. The performance restriction parameters include at least one of the following: entry restrictions, exit restrictions, and applicable license plate range.

[0013] In one embodiment, after the step of outputting the parking package draft after verification, the method further includes: After the parking package draft is manually confirmed and written into the package pool, the transaction result, performance result and refund result corresponding to the parking package draft are obtained; Based on the transaction results, performance results, and refund results, the scoring model and prediction model are updated. Specifically, the actual parking space occupancy data and actual available sales data are used as new training samples to calibrate the reduction coefficient of releaseable capacity and the safety buffer ratio, thereby updating the prediction model.

[0014] In one embodiment, after the step of acquiring multi-source operating data of the target parking lot, the method further includes: The multi-source operational data is anomaly identified, and abnormal data records that do not meet the preset threshold range are removed; After removing outlier data, the operational data from each source are aligned in time series at a uniform time granularity, and missing time slices in the aligned data are interpolated and filled. The data integrity rate of each preset time granularity is calculated based on the data after interpolation and filling. When the data integrity rate is lower than the preset integrity rate threshold, the operating data under the corresponding time granularity is marked as unavailable and a data supplementation command is triggered. The time granularity information marked as unavailable is recorded in the data quality log, which is used to skip data of the corresponding time granularity during subsequent feature extraction; The multi-source business data processed as described above is stored in the cleaned business database.

[0015] In addition, to achieve the above objectives, this application also proposes a parking lot revenue-generating opportunity identification and package generation device, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the parking lot revenue-generating opportunity identification and package generation method described above.

[0016] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the parking lot revenue increase opportunity identification and package generation method described above.

[0017] One or more technical solutions proposed in this application have at least the following technical effects: The technical solution of this application involves acquiring multi-source operating data of a target parking lot, extracting multi-dimensional features from the multi-source operating data, determining the revenue-increasing opportunity level of the target parking lot based on the multi-dimensional features, predicting the distribution of idle time periods and available capacity within a future preset period based on the time-series correlation features in the multi-dimensional features, generating a draft parking package and pre-filling package parameters based on the revenue-increasing opportunity level, the distribution of idle time periods, and the available capacity, performing verification on the draft parking package, the verification including at least time period conflict verification and capacity constraint verification, and outputting the draft parking package after verification.

[0018] The technical solution of this application realizes automatic identification of revenue-generating opportunities and efficient generation of packages. It solves the technical problems of traditional solutions, such as reliance on manual discovery of opportunities, lack of accuracy in prediction, low configuration efficiency, and inability to self-optimize parameters. It realizes a technological leap from manual experience-driven to data intelligence-driven management of parking resource revenue. Attached Figure Description

[0019] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0020] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart illustrating the first embodiment of the parking lot revenue-generating opportunity identification and package generation method of this application. Figure 2 This is a detailed step diagram of step S20 in one embodiment; Figure 3 This is a detailed step diagram of step S30 in one embodiment; Figure 4 This is a detailed step diagram of step S40 in one embodiment; Figure 5 This is a flowchart illustrating the second embodiment of the parking lot revenue-generating opportunity identification and package generation method of the present invention; Figure 6 This is a flowchart illustrating the third embodiment of the parking lot revenue-generating opportunity identification and package generation method of the present invention; Figure 7 This is a schematic diagram of the system structure; Figure 8 This is a schematic diagram illustrating the process of generating a package draft and obtaining manual confirmation. Figure 9This is a schematic diagram of the hardware operating environment involved in the parking lot revenue increase opportunity identification and package generation method in the embodiments of this application.

[0022] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0023] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0024] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0025] In the field of parking resource revenue management, the identification of revenue-generating opportunities and the generation of packages mainly rely on manual experience and static statistical reports. This approach fails to incorporate operational results such as transaction conversion, forecasting deviations, and contract fulfillment anomalies into the feedback loop, making it difficult to balance the efficiency and reliability of large-scale opportunity discovery and refined package generation. When the scale of parking lots expands or the regional competitive landscape changes, the timeliness bottleneck and information blind spots of manual decision-making lead to package pricing deviating from market equilibrium, mismatch between available capacity and idle resources, and a serious lag in responding to revenue-generating opportunities.

[0026] A comprehensive analysis reveals that the core dilemma of the above-mentioned approach lies in the fact that opportunity discovery relies on subjective experience, idle resource analysis lacks accurate prediction, and package configuration relies on manual intervention without a feedback loop. This fundamentally contradicts the inherent needs of the existing asset management stage for scalable efficiency, dynamic adaptability, and parameter self-optimization.

[0027] To address the aforementioned shortcomings, this application proposes a method for identifying parking lot revenue-generating opportunities and generating service packages. This method addresses the core pain points of reliance on manual opportunity identification, insufficient accuracy in idle period prediction, low efficiency in package configuration, and lack of closed-loop optimization. It automates the assessment of revenue-generating opportunity levels through a multi-dimensional weighted scoring model, introduces a time-series prediction model to proactively estimate the distribution of idle time periods and available capacity, automatically generates draft service packages based on a rule engine, and pre-fills in applicable time periods, available capacity, and price ranges. Before outputting to manual confirmation, it enforces checks for time period conflicts, capacity constraints, and price impacts, and feeds back transaction, fulfillment, and refund results to the scoring and prediction models to achieve closed-loop parameter correction.

[0028] Through the above-mentioned technical means, this application realizes the automatic identification of revenue-generating opportunities and the efficient generation of packages, solving the technical problems of traditional solutions such as reliance on manual discovery of opportunities, lack of accuracy in prediction, low configuration efficiency, and inability to self-optimize parameters. It also realizes the technical leap from manual experience-driven to data intelligence-driven management of parking resource revenue.

[0029] Based on this, embodiments of this application provide a method for identifying parking lot revenue-generating opportunities and generating service packages, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the parking lot revenue-generating opportunity identification and package generation method of this application. In this embodiment, the parking lot revenue-generating opportunity identification and package generation method includes steps S10 to S60: Step S10: Obtain multi-source operating data of the target parking lot, and extract multi-dimensional features from the multi-source operating data; In this embodiment, acquiring multi-source operational data of the target parking lot is the data foundation of the entire method. Its purpose is to provide comprehensive and accurate raw data support for subsequent feature extraction, opportunity identification, and package generation. This step requires collecting different types of operational data from multiple data sources and organizing this data in a structured manner for subsequent processing.

[0030] In the specific implementation process, the first step is to establish a data connection with the parking management system of the target parking lot, and obtain entry and exit time-series data through a standardized interface. This entry and exit time-series data includes at least one of the following: historical entry quantity data, historical exit quantity data, parking duration distribution data, and time-period entry frequency data. This data records the specific time points when each vehicle enters and exits the parking lot, as well as the duration of its stay within the parking lot. These raw event data are then statistically summarized according to a preset time granularity, such as aggregating data by hourly or half-hourly time slices, thereby forming continuous time-series data.

[0031] Simultaneously, the system acquires parking space occupancy data for the target parking lot within a preset historical period. This data can be collected in real-time by parking space detection equipment within the parking lot or estimated from entry and exit flow data. Specifically, the system calculates the number of vacant parking spaces at a given moment by subtracting the number of vehicles currently present from the total number of parking spaces, and organizes the occupancy data at the same time granularity as the entry and exit flow data. The acquisition of fee data involves the target parking lot's hourly rental price records, monthly rental price records, and temporary fee records. This data reflects the parking lot's current pricing level and fee structure. Historical transaction data includes historical package transaction records, transaction prices, and transaction time distribution. This data records the transaction status of various package products that have been offered and sold in the past. Regional demand data covers the distribution of surrounding parking resources, surrounding points of interest data, and surrounding event information, such as the vacancy status of other parking lots within a one-kilometer radius, the operating hours of nearby commercial facilities, and whether large-scale events are being held.

[0032] After collecting the aforementioned multi-source operational data, the data is time-series aligned according to a unified time granularity. Since the collection frequency and recording format of different data sources may vary, the system needs to convert all data sources to the same time granularity, for example, aligning them all by hour, to ensure that data from the same time slice can be correlated. The aligned data is then stored in the operational database for subsequent feature extraction steps. This data acquisition and organization process ensures that the data foundation upon which subsequent multi-dimensional feature extraction relies is complete and time-series consistent.

[0033] Step S20: Determine the revenue-increasing opportunity level of the target parking lot based on the multi-dimensional features; To determine the revenue-generating opportunity level of a target parking lot based on multi-dimensional features, a scoring system capable of comprehensively quantifying the parking lot's revenue potential needs to be constructed. Specifically, the extracted multi-dimensional features need to be used as input, and through a series of data processing operations such as normalization, weight configuration, weighted summation, and threshold division, the revenue-generating opportunity level and corresponding parking priority will be output.

[0034] The original feature values ​​for five dimensions—time-based vacancy rate, time-based repetition, regional demand intensity, price sensitivity, and income volatility—were extracted from the multi-dimensional features. The time-based vacancy rate characteristic represents the average proportion of vacant parking spaces within a target time period; a higher vacancy rate indicates a greater abundance of parking spaces available for sale. The time-based repetition characteristic represents the regularity and predictability of vacancy over time; stronger repetition indicates a more stable vacancy pattern within that time period, making it more suitable for commercialization in the form of fixed-time packages. The regional demand intensity characteristic represents the potential demand for parking resources in the surrounding area; this characteristic can be obtained through a comprehensive assessment of the density of surrounding points of interest, pedestrian traffic data, and the supply and demand situation of surrounding parking lots. The price sensitivity characteristic represents the responsiveness of the target user group to price adjustments, which can be estimated through changes in transaction volume before and after historical price adjustments. The income volatility characteristic represents the fluctuation range of existing income and its potential for improvement.

[0035] Since the original feature values ​​of the five dimensions mentioned above have different dimensions and ranges, normalization processing needs to be performed on the original feature values ​​of each dimension before weighted calculation, mapping the values ​​of each dimension to a unified numerical range. Normalization can be performed using extreme value normalization, which uses the minimum and maximum values ​​of the dimension in similar parking lot samples as a benchmark for linear mapping; or using quantile normalization, which uses the quantile of the dimension's value in the sample distribution as the normalization score. Quantile normalization can reduce the interference of outliers on the normalization results. After normalization, the scores of each dimension are all within the same numerical range, such as between 0 and 100 or between 0 and 1, thus eliminating the influence of the difference in dimensions between different dimensions on the subsequent weighted calculation.

[0036] After normalization, weights are assigned to each dimension. The weights can be determined using one or more of the following methods: expert weighting, entropy weighting, or data-driven weighting. Expert weighting, for example, uses the analytic hierarchy process (AHP), where domain experts construct a judgment matrix by comparing each dimension pairwise, and the subjective weights of each dimension are calculated after consistency checks. Entropy weighting determines objective weights based on the information entropy or dispersion of each dimension's indicators on the sample. The greater the dispersion of a dimension's values, the more information that dimension provides for differentiating revenue-generating opportunities in different parking lots, and the higher the weight assigned. Data-driven weighting uses historical revenue generation, package conversion rate, or actual listing success rate as dependent variables, and the normalized scores of each dimension as independent variables. It fits the data using linear regression, logistic regression, or gradient boosting trees, and uses the regression coefficients or feature importance as the weights for each dimension. In some implementations, a combined weighting method can also be used, where the final weights are calculated as a weighted combination of subjective and objective weights, and the combination coefficients can be configured according to the actual application scenario.

[0037] A comprehensive score is calculated by weighted summation of the normalized scores and corresponding weights for each dimension. The comprehensive score is calculated as the sum of the products of the normalized scores for each dimension and their corresponding weights. In some implementations, the system can also introduce interactive correction terms based on the weighted summation, such as the product of the vacancy rate score and the demand intensity score, so that periods with both high vacancy rates and high demand intensity can obtain higher comprehensive scores, thereby more accurately depicting real revenue-generating opportunities.

[0038] After obtaining the comprehensive score, the target parking lot is divided into multiple revenue-generating opportunity levels based on preset thresholds. These levels include high, medium, and low opportunities. The classification can be done using a fixed threshold method: a first and a second threshold are preset; a comprehensive score greater than or equal to the first threshold is considered a high opportunity, a score between the first and second thresholds is considered a medium opportunity, and a score below the second threshold is considered a low opportunity. Alternatively, a dynamic quantile method can be used: among similar parking lots, the top-ranked percentages are classified as high opportunities, the middle percentages as medium opportunities, and the rest as low opportunities. The system can also include a rejection rule: if the normalized score of a key dimension falls below a preset lower limit, the opportunity level is downgraded. For example, if the regional demand score is extremely low, even with a high comprehensive score, the opportunity level of the parking lot will be lowered to avoid generating invalid packages that are difficult to sell.

[0039] The system determines the listing priority based on the revenue-generating opportunity levels identified. High opportunity levels correspond to high listing priority; therefore, package proposals generated by these parking lots are prioritized for inclusion in the package proposal generation queue. Medium opportunity levels correspond to medium listing priority. Low opportunity levels correspond to low listing priority or are temporarily deferred. Through this tiered processing, the system achieves differentiated scheduling of revenue-generating opportunities, ensuring that limited processing resources are prioritized for parking lots with the greatest revenue-generating potential.

[0040] Step S30: Predict the distribution of idle time periods and releaseable capacity within a future preset period based on the time-related features in the multi-dimensional features; In this embodiment, the distribution of idle time periods and available capacity within a preset future period are predicted based on time-series related features among multi-dimensional features, thereby providing quantitative idle resource prediction data for subsequent package generation. This step employs a time-series prediction model, based on historical parking space occupancy data, combined with multiple input variables such as calendar features and supply and demand features, to make forward-looking estimates of the idle status at various time granularities within a future period.

[0041] First, historical parking space occupancy rate time series data for the target parking lot is obtained. This data comes from collected and cleaned parking space occupancy data, organized into a continuous time series according to a uniform time granularity. In addition to the parking space occupancy rate series, calendar features are also needed as auxiliary inputs to the prediction model. Calendar features include at least one of the following: time, day of the week, weekday identifier, weekend identifier, and holiday identifier. These features reflect the periodicity of time itself; for example, parking demand patterns on weekdays and weekends usually differ significantly, and parking demand on holidays may also differ from that on ordinary days. Lag features and moving statistical features are also constructed, including occupancy rates, moving averages, moving variances, and year-on-year and month-on-month changes over previous periods. These features help the model capture short-term fluctuations and trends in parking space occupancy.

[0042] After feature preparation, a suitable time series prediction model is selected based on the data characteristics. The selection criteria include data volume, periodicity intensity, feature richness, and accuracy requirements. When the parking space occupancy data of the target parking lot exhibits strong periodicity and relatively stable patterns, a time series statistical model is chosen, such as a seasonally differentiated autoregressive moving average model or an exponential smoothing model with a seasonal term. These models have low computational cost and strong interpretability. When the input features are multi-dimensional and there are non-linear interactions between features, a gradient boosting model is chosen, such as extreme gradient boosting or a lightweight gradient boosting machine. When the time series is long and there are long-range time dependencies, a recurrent neural network is chosen, such as a long short-term memory network or a gated recurrent unit. In some implementations, multiple models can be combined, such as weighted averaging of the prediction results from multiple models, or a stacked ensemble approach using the outputs of multiple models as input to a meta-model for secondary prediction.

[0043] The prepared feature data is input into the selected prediction model. The model outputs an idle probability sequence or an idle parking space number sequence divided by time granularity for a future preset period. The idle probability reflects the likelihood of a parking lot being idle within a certain time slice, while the idle parking space number directly indicates the expected number of available parking spaces within that time slice. After obtaining the prediction results for each time granularity, the system aggregates consecutive or adjacent idle time slices into idle time period intervals. The purpose of the aggregation operation is to combine scattered time slices into continuous time periods with practical business significance, such as merging multiple consecutive hours into a complete nighttime period. Each idle time period interval obtained after aggregation contains a clear start and end time.

[0044] For each idle time period interval, the corresponding releaseable capacity is determined based on the number of idle parking spaces at each time granularity within that interval. The releaseable capacity can be the average, minimum, or other statistical value of the number of idle parking spaces at each time granularity within that time period; the specific method of value selection can be configured according to actual business needs. In some implementations, both point prediction values ​​and interval prediction values ​​are output simultaneously. For example, a confidence interval for the releaseable capacity is given through quantile prediction, allowing for risk reservation during subsequent conversion of available capacity for package sales. After prediction, the system evaluates the prediction accuracy using evaluation metrics including mean absolute error, root mean square error, and mean absolute percentage error. A preset error threshold can be set; when the prediction error exceeds this threshold, model retraining is triggered to maintain the continuous stability of prediction accuracy.

[0045] Step S40: Based on the revenue increase opportunity level, the idle time distribution, and the release capacity, generate a draft parking package and pre-fill the package parameters; Based on the level of revenue-generating opportunities, the distribution of idle time periods, and available capacity, a draft parking package is generated and pre-filled with package parameters. The purpose is to transform the analysis and prediction results of the previous steps into specific package product plans. This step automatically determines the package type, applicable time period, available capacity, price range, and fulfillment restriction parameters through a series of mapping rules and conversion logic, forming a complete draft parking package.

[0046] The type and priority of the package are determined based on the identified revenue-generating opportunity level. When the revenue-generating opportunity level is high, a draft package is generated first, prioritizing high-value, longer-term packages such as off-peak shared monthly packages or long-term shared packages, which are also assigned a higher listing priority. When the revenue-generating opportunity level is medium, standard package types are generated, such as temporary off-peak multi-user cards or time-sharing packages, assigned a medium listing priority. When the revenue-generating opportunity level is low, conservative or trial packages are generated, such as small-capacity, short-term packages, or the listing is postponed, assigned a lower listing priority.

[0047] After determining the package type, the predicted idle time slots are categorized according to preset time slot types. Preset time slot types include at least one of weekday daytime slots, weekday nighttime slots, weekend slots, and holiday slots. The categorization is based on whether the start and end times of each idle time slot interval match the time range of the preset time slot type. After categorization, the system aggregates consecutive or adjacent idle time slots into applicable time slot intervals for the package, and sets the start and end times of these intervals, as well as entry and exit rules. During the aggregation process, the system removes time slots that overlap with peak business hours to ensure that the applicable time slots for the generated packages do not interfere with the normal temporary parking operations of the parking lot.

[0048] The saleable capacity of the package is calculated based on the releaseable capacity. The conversion method is: releaseable capacity multiplied by a reduction factor, then minus a safety buffer, and rounded down. The reduction factor reflects performance uncertainty and forecasting error; its value is related to the revenue opportunity level. A higher reduction factor can be used for high opportunity levels to fully release resources, while a lower reduction factor is used for low opportunity levels to control risk. The safety buffer ratio is used to reserve parking spaces to meet temporary needs, ensuring that even with some forecasting error, the parking lot still has enough spaces to meet temporary parking demands. The above conversion is performed separately for different time periods; different reduction factors and safety buffer ratios can be configured for the releaseable capacity in different time periods.

[0049] A benchmark price is determined based on the hourly rental price of the target parking lot's main business and the prices of similar parking resources in the surrounding market. Based on the benchmark price, a price adjustment coefficient is calculated according to regional demand characteristics, time-of-day vacancy rate characteristics, and price sensitivity characteristics. The design logic of the price adjustment coefficient is as follows: the higher the demand, the larger the price adjustment coefficient, allowing for a slight increase in the package price; the higher the vacancy rate, the smaller the price adjustment coefficient, allowing for a slight decrease in the package price to boost sales; the higher the price sensitivity, the smaller the price fluctuation range, avoiding customer churn due to excessive price changes. A package price range is generated based on the benchmark price and the price adjustment coefficient. This range includes a minimum and a maximum price, providing an adjustable range for subsequent manual confirmation. Simultaneously, constraints are imposed to ensure that the unit price of the package after conversion is not lower than the cost floor and not higher than the hourly rental price of the main business, preventing price inversions from interfering with the main business.

[0050] Based on the above rules, various parameters for the parking package draft are pre-filled, including applicable time periods, available capacity, price range, and fulfillment restrictions. Fulfillment restrictions include at least one of entry restrictions, exit restrictions, and applicable license plate ranges. Entry and exit restrictions control which time periods package users can enter and leave the parking lot, while the applicable license plate range limits which vehicles can use the package. After the above processing, the system outputs a parking package draft containing complete parameters, which is now ready for sale.

[0051] Step S50: Perform a verification on the draft parking package, the verification including at least time period conflict verification and capacity constraint verification; Verification of the draft parking package is a crucial step in ensuring the rationality and feasibility of the package parameters. Before exporting the draft package to the manual confirmation module, all parameters in the draft must be verified from multiple dimensions, including at least time period conflict verification and capacity constraint verification, to avoid interfering with the normal operation of the parking lot or causing performance risks after the package is launched.

[0052] Time period conflict verification checks whether the applicable time periods of the generated package conflict with the normal operating hours of the parking lot's main business. Specifically, it obtains peak time information for the parking lot's main business, which typically includes weekday morning and evening commuting peaks and weekend shopping and leisure peaks. The applicable time period range in the package draft is compared with the main business peak times to determine if there is any overlap. If the applicable time periods of the package overlap with the main business peak times, it may affect the revenue of the parking lot's normal temporary parking business and may also lead to competition for parking spaces between package users and temporary parking users. For package drafts with time period conflicts, the applicable time periods of the package can be automatically adjusted to avoid the main business peak times, or the conflict information can be marked for reference during subsequent manual confirmation. In some implementations, time period conflict verification also includes conflict checks between different time periods within the package, such as whether the applicable time periods of different packages within the same parking lot overlap, to avoid multiple packages competing for limited parking space resources during the same time period.

[0053] Capacity constraint verification checks whether the available capacity of a package exceeds the actual carrying capacity of the parking lot. It obtains the total number of parking spaces in the target parking lot and the predicted occupancy rate for each time period. Combined with the predicted release capacity, it determines whether the available capacity set in the package draft is reasonable. Specifically, it compares the package's available capacity with the release capacity for the corresponding time period. If the available capacity is greater than the release capacity, it means that the number of parking spaces promised by the package exceeds the actual number of idle parking spaces the parking lot can provide during that time period. This may lead to the risk of package users not finding available parking spaces upon arrival. In cases where the available capacity exceeds the release capacity, the available capacity is automatically reduced to within the release capacity range. Capacity constraint verification also includes checking the cumulative available capacity of multiple packages. If the sum of the available capacity of multiple packages in the same time period exceeds the release capacity for that time period, the system needs to allocate or adjust the capacity according to the priority of each package.

[0054] In addition to time-slot conflict and capacity constraint checks, the system can also perform main business price impact checks and performance risk checks. The main business price impact check examines whether package prices will interfere with the parking lot's main business pricing system. For example, it checks whether the package's converted unit price is lower than the cost floor or excessively lower than the main business's hourly rental price, preventing users from switching heavily from temporary parking to package services due to excessively low package prices, thus reducing overall revenue. The performance risk check examines whether the package's performance restriction parameters are reasonable and feasible. For example, it checks whether entry and exit restrictions are consistent with the parking lot's actual operating rules and whether the format of the applicable license plate range is correct.

[0055] During the verification process, if any verification fails, the specific reason for the failure is recorded, and the draft package is attempted to be automatically corrected according to the preset correction rules. If the automatic correction still fails, the draft package is marked as pending manual processing, and the specific information of the verification failure is output to the manual confirmation module. Only draft packages that pass all verifications can proceed to the subsequent manual confirmation stage.

[0056] Step S60: Output the draft parking package after verification.

[0057] Outputting the approved parking package draft to the manual confirmation module is the connecting link between the automated package generation process and human decision-making. This step is responsible for presenting the package draft generated by all previous processes in a structured manner to authorized operators, who will then conduct a final review of the draft content and decide whether to add it to the package pool.

[0058] First, the approved parking package drafts are organized into a structured confirmation interface. This structured confirmation interface includes at least a complete display of the package draft, explicit presentation of key parameters, identification of editable fields, and three operation entry points: confirm, modify, or reject. The complete display of the package draft includes the package type, applicable time period, available capacity, upper and lower limits of the price range, and various performance restrictions. Key parameters are presented in a prominent manner; for example, applicable time periods, prices, and available capacity are displayed in a prominent manner in the main area of ​​the interface, allowing operators to quickly access core information. Editable fields allow operators to adjust certain parameters of the package draft before confirmation, such as fine-tuning the upper or lower limit of the price range or modifying the available capacity. The system will perform real-time verification of the operator's modifications to ensure that the modified parameters still meet all constraints.

[0059] Along with the draft package proposal, information explaining its creation will be displayed. This information includes the parking lot's revenue-generating opportunity level and its overall score, detailed normalized scores for each dimension, confidence levels for idle time predictions, and verification results from the validation process. This information helps operators understand the logic behind the package proposal's creation and parameter settings, providing ample decision-making reference for manual assessment. The scope of operable fields and the types of operations that can be performed will be controlled based on the operator's permission level; operators with different permission levels may have different modification and confirmation permissions.

[0060] After reviewing the draft package through the structured confirmation interface, operators can choose one of three actions: confirm, modify, or reject. If the operator confirms, the draft package is written to the package pool, and packages in the pool will enter the formal listing and sales process. If the operator modifies, the system receives the submitted changes, re-validates the modified package parameters, and writes the modified draft package to the package pool after successful validation. If the operator rejects, the reason for rejection is recorded, and the draft package is archived in the history, not written to the package pool. All operation records are fully logged by the system, including operator information, operation time, operation type, and parameter comparisons before and after modification. This log data can serve as a reference for subsequent auditing and model optimization.

[0061] Once the draft package is written into the package pool, it is pushed to the parking lot's sales channels for sale. The sales and fulfillment of the package are continuously tracked, and feedback data such as transaction results, fulfillment results, and refund results are collected. Transaction results include information such as whether a transaction was completed, the percentage of available capacity, conversion rate, and transaction price. Fulfillment results include data such as actual usage rate, fulfillment rate, time slot conflict rate, and abnormal rates such as overdue parking. Refund results include refund rate, refund amount percentage, and distribution of refund reasons. This feedback data is used for subsequent model updates, achieving a complete data loop from opportunity identification and package generation to effect feedback.

[0062] Further, see Figure 2 , Figure 2 This is a detailed step diagram of step S20 in one embodiment, based on the shown... Figure 2 The step of determining the revenue-generating opportunity level of the target parking lot based on the multi-dimensional features includes S21~S25: Step S21: Normalize each dimension of the multi-dimensional features to obtain the normalized score of each dimension. The multi-dimensional features include time period vacancy rate feature, time period repetition feature, regional demand heat feature, price sensitivity feature and income fluctuation feature. Step S22: Assign weights to each dimension, wherein the weights are determined by at least one of the following methods: expert weighting, entropy weighting, or data-driven weighting. Step S23: Calculate the comprehensive score by weighting and summing the normalized scores of each dimension with their corresponding weights. Step S24: Based on the comprehensive score and the preset classification threshold, the target parking lot is divided into multiple revenue-increasing opportunity levels, wherein the revenue-increasing opportunity levels include high opportunity, medium opportunity and low opportunity. Step S25: Determine the corresponding listing priority based on the revenue increase opportunity level obtained from the classification. The high opportunity corresponds to the high listing priority, the medium opportunity corresponds to the medium listing priority, and the low opportunity corresponds to the low listing priority or postponement of listing.

[0063] This embodiment defines a complete automated assessment process for revenue-generating opportunity levels. The process begins with the normalization of multi-dimensional features, followed by weight configuration, weighted summation, and threshold division, ultimately outputting the revenue-generating opportunity level and its corresponding listing priority. This series of operations transforms the raw feature values ​​extracted in previous steps into level determination results with clear business implications, providing a priority decision basis for subsequent package generation.

[0064] First, we obtain the five dimensions of features extracted from the multi-source operational data in the previous steps. These five dimensions are: time-period vacancy rate, time-period repetition, regional demand intensity, price sensitivity, and income fluctuation. Since the original feature values ​​of these five dimensions have completely different dimensions and ranges—for example, the vacancy rate might be expressed as a ratio between 0 and 1, regional demand intensity might be expressed as a normalized composite index, and price sensitivity might be expressed as the rate of change in transaction volume before and after historical price adjustments—these data with different dimensions cannot be directly compared using weighted averages. Therefore, it is necessary to perform normalization on the original feature values ​​of each dimension separately, mapping the values ​​of each dimension to a unified numerical range. After normalization, the scores of each dimension are all within the same numerical range between 0 and 100 or between 0 and 1, thus eliminating the unfair impact of the difference in dimensions between different dimensions on subsequent weighted calculations.

[0065] The specific normalization method can be flexibly selected based on the actual data distribution characteristics. Extreme value normalization uses the minimum and maximum values ​​of that dimension among similar parking lot samples as a benchmark for linear mapping, converting the original value into its relative position within the range of values ​​for that dimension. Quantile normalization uses the quantile position of the dimension's value in the sample distribution as the normalization score. The advantage of this method is that it reduces the interference of extreme outliers on the normalization results. When a parking lot has particularly extreme values ​​in a certain dimension, quantile normalization will not distort the normalization results for other normal values ​​due to this one extreme value. Alternatively, it can automatically select a more suitable normalization method based on the actual data distribution, or provide both normalization results simultaneously for subsequent steps.

[0066] Weights are assigned to each normalized dimension, with the physical meaning of the weight being the importance of each dimension in the overall score. Weighting methods can include expert weighting, such as using the analytic hierarchy process (AHP) where domain experts construct judgment matrices by comparing each dimension pairwise. Experts use their practical business experience to determine whether vacancy rate or demand intensity contributes more to revenue growth opportunities, or whether revenue volatility or price sensitivity has a stronger ability to distinguish revenue growth opportunities. After consistency testing, the subjective weights of each dimension are calculated. Entropy weighting determines objective weights based on the information entropy or dispersion of each dimension's indicators across the sample. The greater the dispersion of a dimension's values ​​across a large number of parking lot samples, the stronger its ability to distinguish different parking lots, the greater the amount of information it contains, and therefore the higher its weight. Data-driven weighting uses historical revenue growth, package conversion rate, or actual listing success rate as dependent variables, and the normalized scores of each dimension as independent variables. Linear regression or gradient boosting trees are used for fitting, and the regression coefficients or feature importance are used as the weights of each dimension. In some implementations, a combined weighting method can also be used, in which subjective weights and objective weights are weighted and combined to obtain the final weight. The combination coefficient can be configured according to the actual application scenario.

[0067] The comprehensive score is calculated by weighting and summing the normalized scores of each dimension with their corresponding weights. The formula for the comprehensive score is the sum of the products of the normalized scores of each dimension and their corresponding weights. For example, assuming the normalized vacancy rate score is 80, the time period repetition score is 70, the regional demand heat score is 90, the price sensitivity score is 60, and the income volatility score is 75, with corresponding weights of 0.3, 0.2, 0.25, 0.15, and 0.1 respectively, the comprehensive score would be 80 multiplied by 0.3, plus 70 multiplied by 0.2, plus 90 multiplied by 0.25, plus 60 multiplied by 0.15, plus 75 multiplied by 0.1, resulting in a score of 78.5. In some implementations, interactive correction terms can be introduced on the basis of weighted summation. For example, the product of vacancy rate score and demand heat score can be introduced as a correction term and added to the comprehensive score. In this way, parking lots with both high vacancy rate and high demand heat can obtain a higher comprehensive score. This design more accurately describes the judgment logic of real revenue increase opportunities, that is, revenue increase opportunities are most significant only when resource supply is sufficient and market demand is strong at the same time.

[0068] Based on the comprehensive score and preset thresholds, target parking lots are divided into multiple revenue-increasing opportunity levels. These levels include high, medium, and low opportunities. One method is a fixed threshold method, with preset first and second thresholds, where the first threshold is greater than the second. A comprehensive score greater than or equal to the first threshold is considered a high opportunity; a score between the first and second thresholds is considered a medium opportunity; and a score below the second threshold is considered a low opportunity. For example, if the first threshold is 75 points and the second threshold is 50 points, parking lots with a comprehensive score of 75 or higher are classified as high-opportunity, those between 50 and 75 points as medium-opportunity, and those below 50 points as low-opportunity. Another method is a dynamic quantile method. In this method, the top 25% of parking lots in the comprehensive score distribution of similar parking lots are classified as high-opportunity, those between 25% and 60% as medium-opportunity, and the rest as low-opportunity. The thresholds for this method are dynamically updated based on the sample distribution.

[0069] In the classification logic, a veto rule can also be set as a supplementary mechanism for level determination. The veto rule means that when the normalized score of a certain key dimension falls below a preset lower limit, even if the overall score meets the standard for a higher level, the opportunity level of that parking lot will be downgraded by one or more levels. For example, regional demand is one of the key dimensions for judging revenue growth opportunities. When a parking lot's regional demand score is extremely low, it indicates a lack of sufficient parking demand in the surrounding area. Even if the parking lot itself has a high vacancy rate and a regular vacancy pattern, it will be difficult to generate sales for packages under such demand conditions. Therefore, the opportunity level of that parking lot will be downgraded from high or medium opportunity to a lower level. The veto rule avoids the situation where the overall score masks serious deficiencies in a single dimension, making the opportunity level determination results closer to actual business scenarios.

[0070] The system determines the listing priority based on the revenue-generating opportunity level. The listing priority determines the order in which the package proposals generated by a parking lot enter the package proposal generation queue. High opportunity levels correspond to high listing priority; package proposals from these parking lots enter the package proposal generation queue first, and the system allocates more processing resources to them. Medium opportunity levels correspond to medium listing priority. Low opportunity levels correspond to low listing priority or delayed listing. Through this hierarchical scheduling mechanism, the system ensures that processing resources are prioritized for parking lots with the greatest revenue-generating potential, achieving efficient integration between revenue-generating opportunity identification and package generation.

[0071] Furthermore, after the step of dividing the target parking lot into multiple revenue-generating opportunity levels based on the comprehensive score and a preset classification threshold, the method further includes: Obtain the normalized score of at least one key dimension of the target parking lot; When the normalized score of the key dimension is lower than the preset lower limit, the income increase opportunity level will be downgraded by one or more levels. The key dimension includes at least the regional demand heat feature.

[0072] In this embodiment, after classifying the target parking lot into high-opportunity, medium-opportunity, or low-opportunity levels based on the comprehensive score and preset classification thresholds, a supplementary verification step is required. The core function of this step is to review the reasonableness of the already assessed opportunity levels, ensuring that no serious deficiencies in any key dimension are masked by the comprehensive score. The comprehensive score itself is a weighted aggregation result, which in most cases can reflect the overall revenue potential of the parking lot well. However, in some special scenarios, the parking lot may perform well in most dimensions but have a fatal weakness in a crucial dimension. In this case, the weighted aggregation score may still reach the standard of a high level, resulting in a level determination that does not conform to the actual situation. To address this situation, a rejection rule is introduced as a post-verification mechanism for opportunity level classification.

[0073] The execution logic of the veto rule is that the system obtains the normalized score of at least one key dimension from the calculated normalized scores of each dimension. Among these key dimensions, the regional demand heat characteristic is listed as a mandatory key dimension. Regional demand heat reflects the potential demand intensity for parking resources in the area surrounding the parking lot. The score of this dimension directly determines whether sufficient user willingness to purchase can be obtained even after the package is generated and put on the market. If a parking lot has an extremely low regional demand heat score, it means that the parking demand in the area where the parking lot is located is very weak. Even if the parking lot has a large number of idle parking spaces and the vacancy pattern is very regular, without demand support, it is difficult to obtain an ideal market response for any package generated, and there may even be a situation where no one inquires about the package after it is put on the market.

[0074] The normalized scores of the key dimensions are compared with preset lower limits. These preset lower limits are threshold parameters configured independently for each key dimension. Their specific values ​​can be set based on business experience or determined by analyzing the score distribution of successful and unsuccessful packages across each dimension in historical data. Taking the regional demand heat dimension as an example, assuming the preset lower limit is set to 30 points, when a parking lot rated as high-opportunity has a normalized score of only 25 points in the regional demand heat dimension, it is determined that although the parking lot's overall score meets the high-opportunity standard, its regional demand heat is severely insufficient. In this case, a rejection rule is triggered to downgrade the parking lot's opportunity level.

[0075] The downgrade operation can be performed using a step-by-step downgrade or a skip-level downgrade. A step-by-step downgrade means the system lowers the opportunity level by one level, for example, from high opportunity to medium opportunity, or from medium opportunity to low opportunity. A skip-level downgrade means the system directly lowers the opportunity level to the lowest possible level, for example, from high opportunity to low opportunity. The specific downgrade amount can be determined based on the degree of deviation of the key dimension score from a preset lower limit; the greater the deviation, the larger the downgrade. In some implementations, the system can also set multiple lower limits to form a tiered judgment logic; for example, when the key dimension score is below the first lower limit, it is downgraded by one level, and when it is below the second lower limit, it is downgraded by two levels. After the downgrade operation is completed, the reason for the downgrade and the specific veto rule triggered are recorded. This information can later be output to the manual confirmation module as part of the basis for generating the package draft, helping operators understand the reason for the level adjustment.

[0076] During the execution of the veto rules, the basic principle of grade downgrading must be maintained: the final grade after downgrading should not be lower than the lowest grade. Furthermore, if multiple veto rules are triggered for the same parking lot, the system will execute the downgrading according to the maximum reduction specified in each rule. Through supplementary verification of the veto rules, the system avoids situations where comprehensive scoring masks serious deficiencies in a single dimension, making the determination of revenue-generating opportunity grades more closely aligned with the feasibility requirements of actual business scenarios and improving the reliability of the entire method in actual operation.

[0077] Further, see Figure 3 , Figure 3 This is a detailed step diagram of step S30 in one embodiment, based on the shown... Figure 3 The step of predicting the distribution of idle time periods and releaseable capacity within a future preset period based on the time-series correlation features in the multi-dimensional features includes S31~S34: Step S31: Obtain historical parking space occupancy rate time series data of the target parking lot; Step S32: Based on the historical parking space occupancy rate time series data and calendar features, predict the idle probability or number of idle parking spaces at each time granularity within a future preset period using a time series prediction model. Step S33: Aggregate consecutive or adjacent idle time slices into idle time intervals, each of the idle time intervals including start and end times; Step S34: Determine the corresponding release capacity based on the number of idle parking spaces in each idle time period interval. The calendar features include at least one of time, day of the week, weekday identifier, weekend identifier, and holiday identifier.

[0078] This embodiment defines a complete idle time period prediction process. This process begins with acquiring historical parking space occupancy data, proceeds through inference calculations using a time-series prediction model, and ultimately outputs idle time period intervals with clearly defined start and end times, along with the corresponding available capacity. This series of operations provides a quantitative data foundation of idle resources for subsequent package generation, ensuring that the applicable time periods and available capacity of the proposed packages are based on verifiable data.

[0079] Obtain historical parking space occupancy rate time series data for the target parking lot. This data comes from collected and cleaned parking space occupancy data. Organize the parking space occupancy rate series according to a uniform time granularity, which can be set to hourly or half-hourly based on actual prediction needs. The time span of the historical data should be sufficient to cover multiple complete business cycles, such as including data from at least the past three months or even longer, to fully capture different occupancy patterns between weekdays and weekends, between weekdays, and between holidays and ordinary days. During the data acquisition process, the continuity and completeness of the data should be checked. If data at a certain time granularity is missing or abnormal, preprocessing can be performed according to the data cleaning and imputation methods described above to ensure that the data quality input into the prediction model meets the requirements.

[0080] Historical parking space occupancy time-series data and calendar features are input into a time-series prediction model. The model outputs the probability of vacancy or the number of vacant parking spaces at each time granularity within a preset future period. Calendar features are key input variables to aid prediction, helping the model understand the cyclical patterns of time itself. Calendar features specifically include at least one of the following: time of day, day of the week, weekday identifier, weekend identifier, and holiday identifier. Time of day features reflect the inherent differences in parking demand at different times of the day; for example, 8:00 AM and 2:00 PM typically have completely different parking space occupancy situations. Day of the week features reflect cyclical changes within a week; for example, there are systematic differences in parking demand patterns between Monday to Friday and Saturday and Sunday. Weekday and weekend identifiers further abstract the day of the week into two date types, making it easier for the model to capture the overall differences between these two types of dates. Holiday identifiers mark statutory holidays and adjusted workdays, which have special demand patterns that differ from ordinary weekdays or weekends. In addition to calendar features, the system can also construct lag features and sliding statistical features as supplementary inputs, such as the occupancy rate of the same type of date in the previous period and the average occupancy rate of the same time granularity in the past week. These features help the model to more accurately grasp the short-term fluctuation pattern of parking space occupancy.

[0081] The specific selection of the time series prediction model can be flexibly configured based on the data characteristics. When the parking space occupancy data of the target parking lot exhibits strong periodicity and relatively stable patterns, a time series statistical model can be selected, such as a seasonally differentiated autoregressive moving average model or an exponential smoothing model with a seasonal term. These models have lower computational overhead and better interpretability of prediction results. When the input features are rich in dimensions and there are nonlinear interactions between features, a gradient boosting model can be selected, such as extreme gradient boosting or a lightweight gradient boosting machine. These models can automatically learn complex relationships between features. When the time series is long and there are long-range time dependencies, a recurrent neural network can be selected, such as a long short-term memory network or a gated recurrent unit. In some implementations, the system can also use a combination of multiple models, i.e., first use multiple models to make predictions separately, and then perform a weighted average or stacked ensemble of the prediction results of each model. The combination weights are determined based on the prediction errors of each model on the validation set. The model training process adopts a time-series cross-validation method, dividing the training set, validation set, and test set in chronological order to avoid interference from future information leakage on model evaluation. Model hyperparameters are tuned using grid search or Bayesian optimization, with validation set error as the selection criterion.

[0082] The idle time slices from the prediction results at each time granularity are aggregated. First, it's determined whether each time granularity is idle or not. For models that output idle probability, the system sets a probability threshold; when the idle probability of a time granularity exceeds this threshold, it's considered idle. For models that directly output the number of idle parking spaces, a parking space number threshold is set; when the number of idle parking spaces at a time granularity exceeds this threshold, it's considered idle. After determining the status, all time slices determined to be idle within the entire prediction period are traversed, and consecutive or adjacent idle time slices are merged into a consecutive idle time interval. Adjacent means that although there is an interval between two time slices, the length of the interval is within the allowable range. For example, if two idle time slices are separated only by a short non-idle time slice, then business rules can be used to decide whether to merge these two idle time slices into the same idle time interval. Each aggregated idle time slot includes a clear start and end time, such as 10 PM to 6 AM the following day on a certain day of a month, or 9 AM to 5 PM on a certain Saturday. These aggregated idle time slots have direct business meaning and can be directly mapped to the applicable time periods of the package.

[0083] The system determines the corresponding release capacity based on the number of idle parking spaces within each idle time period interval. For each aggregated idle time period interval, the predicted number of idle parking spaces at each time granularity within that interval is obtained. The release capacity can be determined by taking the average, minimum, or specific quantile value of the number of idle parking spaces at each time granularity within that time period. The specific method can be configured according to business needs and risk preferences. A more conservative approach is to take the minimum or lower quantile value of the number of idle parking spaces within that time period as the release capacity, which ensures that there are enough parking spaces available for package users throughout the entire applicable period of the package. A more aggressive approach is to take the average number of idle parking spaces within that time period as the release capacity. The system can also further reserve risk based on the release capacity, for example, by deducting a certain percentage of safety buffer from the release capacity to cope with the risks brought about by prediction uncertainty. After the prediction is completed, the system evaluates the prediction accuracy using evaluation indicators including mean absolute error, root mean square error, and mean absolute percentage error. For the determination of idle and non-idle spaces, accuracy, precision, recall, and F1 score are also used for evaluation. A preset error threshold is set. When the prediction error exceeds the threshold, the model is retrained to maintain the continuous stability of prediction accuracy.

[0084] Regarding the above Figure 3 Step S32 is further refined, namely, the step of predicting the idle probability or number of idle parking spaces at each time granularity within a future preset period based on the historical parking space occupancy rate time series data and calendar features using a time series prediction model, includes S32-1 to S32-2: Step S32-1: Obtain the regional demand data, surrounding points of interest data, and competitor parking lot vacancy data for the target parking lot; Step S32-2: Input the regional demand data, the surrounding points of interest data, and the competitor parking lot vacancy data as exogenous variables into the time series prediction model. The time series prediction model is any one of a time series statistical model, a gradient boosting model, or a recurrent neural network. When the parking space occupancy data of the target parking lot shows periodicity and stable regularity, the time series statistical model is selected. When the input feature dimension contains exogenous variables and there is a non-linear interaction relationship between features, the gradient boosting model is selected. When the time series has a long-range dependency relationship and the sample size is sufficient, the recurrent neural network is selected.

[0085] This embodiment further refines the input features and model selection logic of the time-series prediction model. Based on historical parking space occupancy time-series data and calendar features, it introduces regional demand data, surrounding point-of-interest data, and competitor parking lot vacancy data as exogenous variables, and adaptively selects the most suitable prediction model type according to the data characteristics of the target parking lot. This enables the prediction model to perceive changes in the external supply and demand environment of the parking lot, while ensuring reliable prediction accuracy under different data conditions.

[0086] Three types of exogenous data are acquired for the target parking lot. The first type is regional demand data, which reflects the overall intensity of parking demand within the area where the target parking lot is located. Regional demand data can be obtained in various ways, such as vehicle ownership data from city traffic management departments, navigation destination search frequency data from navigation map service providers, or population density distribution data from mobile operators. Regional demand data is usually organized at the same time granularity as the parking space occupancy rate series so that it can be used as a time-aligned exogenous variable input into the prediction model in subsequent steps. The second type is surrounding point-of-interest (POI) data, which describes the distribution of various facilities within a certain range around the target parking lot. Different types of POIs affect parking demand in different ways. For example, restaurant POIs generate a large amount of short-term parking demand during mealtimes, shopping POIs significantly increase parking demand on weekends and holidays, while office POIs are the main source of parking demand during weekday daytime hours. When acquiring surrounding points of interest (POIs) data, the system typically collects data centered on the target parking lot within a preset buffer radius. This buffer radius can be configured based on the land use type and facility density of the parking lot's location; for example, the buffer radius can be smaller in urban centers and larger in suburban areas. The system can also count different categories of POIs separately, forming multi-dimensional POI feature vectors. These feature vectors can more precisely characterize the impact of the surrounding environment on parking demand. The third category is competitor parking lot vacancy data. This data reflects the parking space occupancy status of other parking lots surrounding the target parking lot during the same time period. There is usually a substitution or linkage relationship between the vacancy rates of competitor parking lots and the target parking lot. For example, when surrounding competitor parking lots are full, some vehicles may turn to the target parking lot, thus reducing its vacancy rate. By acquiring real-time or historical vacancy data from competitor parking lots, the system can capture this competitive substitution effect, thereby improving the accuracy of predicting the target parking lot's idle time periods. The selection range of competitor parking lots can be determined based on the spatial distance between parking lots and user parking behavior characteristics.

[0087] The acquired regional demand data, surrounding points of interest data, and competitor parking lot vacancy data are used as exogenous variables, which, along with historical parking space occupancy time series data and calendar features, are input into the time series prediction model. The introduction of exogenous variables allows the prediction model to not only rely on the parking lot's own historical occupancy patterns but also to perceive changes in the external environment. For example, when a large-scale event is held in the surrounding area, leading to a significant increase in regional demand, the model can capture this change through exogenous variables and adjust the prediction results accordingly. The specific type of time series prediction model can be selected from three options: time series statistical models, gradient boosting models, and recurrent neural networks. Different types of models have different applicable scenarios and data requirements. The system adaptively selects the most suitable model type based on the data characteristics of the target parking lot.

[0088] When the parking space occupancy data of the target parking lot exhibits strong periodicity and relatively stable patterns, the system selects a time series statistical model. Time series statistical models are represented by the seasonal difference autoregressive moving average model and the exponential smoothing model with a seasonal term. The core assumption of these models is that future values ​​of the time series can be linearly predicted from their historical values ​​and seasonal components. In parking lots where parking space occupancy data has obvious daily and weekly cycles and does not exhibit large random fluctuations, time series statistical models can achieve good predictive results with lower computational overhead. Furthermore, their prediction results are highly interpretable, making it easy for operators to understand the basis of the predictions. Time series statistical models do not rely on complex interactions between exogenous variables; therefore, when the relationship between exogenous variables and parking space occupancy rates is primarily linear, these models are an efficient and reliable choice.

[0089] When the input feature dimension includes exogenous variables and there are non-linear interactions between features, the system selects a gradient boosting model. Gradient boosting models, represented by extreme gradient boosting and lightweight gradient boosting machines, integrate multiple weak learners to progressively fit the residuals, automatically learning non-linear interactions and high-order combined features between features. In parking space occupancy prediction scenarios, there may be an interaction effect between regional demand and the distribution of surrounding points of interest. For example, in areas with a high concentration of restaurant interests, the impact of regional demand on parking space occupancy is much greater during mealtimes than outside of mealtimes. This interaction effect is difficult to characterize with a linear model, but the gradient boosting model can automatically capture such complex non-linear relationships through the hierarchical splitting structure of decision trees. The gradient boosting model can also output feature importance rankings, helping operators identify which exogenous variables contribute the most to the prediction results.

[0090] Recurrent neural networks (RNNs) are chosen when long-range dependencies exist in time series data and the sample size is sufficient. RNNs, represented by Long Short-Term Memory (LSTM) networks and gated recurrent units (GRUs), effectively capture long-range dependencies in time series data by introducing gating mechanisms and recurrent connection structures. In parking space occupancy prediction, some influencing factors may have long lag effects. For example, the impact of a large event on parking demand may be apparent hours before the event begins, or the impact of a policy change on parking behavior may last for weeks. RNNs can retain information from distant time steps through their internal memory units, thus fully utilizing long-range dependencies in historical data when predicting current parking space occupancy rates. Training RNNs requires a large amount of sample data to avoid overfitting; therefore, when the target parking lot has accumulated a sufficiently long period of historical data, these models can achieve high prediction accuracy.

[0091] In some implementations, multiple model combinations can be used to further enhance prediction robustness. For example, time-series statistical models, gradient boosting models, and recurrent neural networks can be used simultaneously for prediction. Then, the prediction results of each model are weighted and averaged or stacked and integrated, with the combined weights dynamically determined based on the prediction errors of each model on the validation set. This combination approach can integrate the advantages of different models and maintain stable prediction performance even when a single model may fail. After model training, time-series cross-validation is used to evaluate prediction accuracy. Evaluation metrics include mean absolute error, root mean square error, and mean absolute percentage error. When the prediction error exceeds a preset threshold, model retraining or parameter adjustment is triggered.

[0092] Further, see Figure 4 , Figure 4 This is a detailed step diagram of step S40 in one embodiment, based on the shown... Figure 4 The step of generating a draft parking package and pre-filling package parameters based on the revenue increase opportunity level, the idle time distribution, and the releaseable capacity includes S41~S45: Step S41: Determine the package type and package priority based on the revenue increase opportunity level; Step S42: Classify the idle time period distribution according to the preset time period type and map it to at least one of weekday daytime time period, weekday nighttime time period, weekend time period or holiday time period, and aggregate consecutive or adjacent idle time periods into the time period range applicable to the package. Step S43: Calculate the saleable capacity of the package based on the releaseable capacity. The saleable capacity of the package is obtained by multiplying the releaseable capacity by a reduction factor and deducting the safety buffer, then rounding down. Step S44: Determine the benchmark price based on the hourly rental price of the target parking lot's main business and the prices of similar parking resources in the surrounding market, and calculate the price adjustment coefficient based on the regional demand characteristics, the time period vacancy rate characteristics, and the price sensitivity characteristics. Generate a package price range based on the benchmark price and the price adjustment coefficient. Step S45: Pre-fill the applicable time period, available capacity, price range and performance restriction parameters of the parking package draft. The performance restriction parameters include at least one of the following: entry restrictions, exit restrictions and applicable license plate range.

[0093] This embodiment defines a complete automated process for generating parking package drafts. It takes three key pieces of information—revenue-generating opportunity level, idle time distribution, and available capacity—output from previous steps as input. Through a series of mapping rules, conversion logic, and calculation methods, it automatically generates a parking package draft containing complete fields such as package type, applicable time period, available capacity, price range, and fulfillment restriction parameters. The core design principle of this series of operations is to transform abstract level determination and prediction results into package product solutions with clear business meaning that can be directly put on the market for sale.

[0094] The type and priority of packages are determined based on the identified revenue-generating opportunity level. Different revenue-generating opportunity levels represent different levels of revenue potential for parking lots, therefore, different types of package products should be matched to achieve differentiated business strategies. When the revenue-generating opportunity level is high, it indicates that the parking lot has ample idle resources and strong regional demand during the target period. In this case, priority should be given to generating package drafts and focusing on high-value, long-term package types. High-value, long-term packages, such as off-peak shared monthly packages or long-term shared packages, have higher unit prices and longer usage periods, which can bring stable and considerable revenue to the parking lot. At the same time, long-term packages are conducive to cultivating user habits and loyalty, forming continuous repeat purchase behavior. At the same time, high-opportunity levels are assigned high listing priority to ensure that these package drafts with the greatest revenue potential enter the generation queue and subsequent manual confirmation process first. When the revenue-generating opportunity level is medium, it indicates that the parking lot has the conditions for increasing revenue, but not as sufficient as the high-opportunity level. In this case, regular package types, such as temporary off-peak pass cards or time-sharing packages, should be generated. The characteristics of standard packages are high flexibility and low barriers to entry. Users can purchase and use them flexibly according to their own needs. These packages are suitable for scenarios with moderate revenue growth potential but still requiring market validation. Medium opportunity levels are assigned medium listing priority. When the revenue growth opportunity level is low, it indicates that there is significant uncertainty regarding the idle resources or regional demand of the parking lot. In this case, conservative or trial packages are generated, such as small-capacity packages with short cycles. Small capacity means fewer available quantities to control risk, and short cycles mean a shorter validity period for quick market response validation. Low opportunity levels correspond to low listing priority or delayed listing. Delayed listing means that although the system has generated a package draft, it will not immediately enter the manual confirmation queue but will be reserved for processing when conditions are ripe. Through this tiered package type mapping mechanism, accurate matching between revenue growth opportunity identification results and package product solutions is achieved.

[0095] The predicted idle time slots are categorized according to preset time slot types, and then aggregated into applicable time slot intervals for the parking packages. Preset time slot types are business-oriented divisions of parking lot operating hours, including at least one of the following: weekday daytime, weekday nighttime, weekend, and holiday. Weekday daytime typically covers the morning to afternoon of weekdays, catering to the parking needs of office workers and those conducting business. Weekday nighttime typically covers the evening to the following morning, catering to the nighttime parking needs of nearby residents. Weekend time slots cover the entire Saturday and Sunday, catering to the parking needs of leisure shopping and family outings. Holiday time slots cover statutory holidays and their adjusted workdays, exhibiting unique demand patterns different from regular weekdays and weekends. The system matches each predicted idle time slot interval with the time range of the aforementioned preset time slot types to determine which preset time slot type each idle time slot interval belongs to. It can also perform multi-label categorization based on the degree of overlap between the predicted start and end times of the idle time slot intervals and the time range of the preset time slot types.

[0096] After classifying the time periods, consecutive or adjacent idle time periods are further aggregated into applicable time periods for the packages. Specifically, if multiple adjacent idle time periods exist under a certain preset time period type, and the intervals between these intervals are small, these intervals are merged into a larger continuous interval. The start and end times of the merged interval are the applicable time periods for the package. During the aggregation process, time periods that overlap with the main business peaks must be removed to ensure that the applicable time periods for the generated packages do not interfere with the normal temporary parking revenue of the parking lot. The main business peak periods can be identified based on historical data, such as weekday morning and evening commuting peaks and weekend shopping peaks. If the predicted idle time period intervals overlap with the main business peak periods, the overlapping parts are removed from the applicable time periods of the packages to ensure that package sales do not affect the normal temporary parking business. After classification, aggregation, and removal operations, each package draft has obtained a clear and business-reasonable applicable time period range.

[0097] The saleable capacity of the package is calculated based on the release capacity. Release capacity represents the predicted number of idle parking spaces, reflecting the total amount of idle resources a parking lot can theoretically provide during the target period. The saleable capacity of the package, however, is the actual quantity that can be sold. Therefore, the release capacity needs to be converted and adjusted. The conversion method is to multiply the release capacity by a reduction factor, deduct a safety buffer, and then round down. The reduction factor is introduced to reflect the impact of performance uncertainty and prediction errors. Although the prediction model can provide relatively accurate estimates of idle time distribution and release capacity, various unforeseen factors in actual operation can cause deviations between predicted and actual values. The reduction factor serves to provide a safety margin for such uncertainty. The value of the reduction factor is related to the revenue opportunity level. A higher reduction factor can be used for high opportunity levels, meaning the system believes the prediction results for high opportunity level parking lots are more reliable, thus allowing for more full release of idle resources for sale. A lower reduction factor is used for low opportunity levels, meaning the system has doubts about the reliability of the prediction results and therefore adopts a more conservative saleable capacity conversion strategy to control risk. The safety buffer ratio is an additional reserve space beyond the reduction factor, designed to ensure that the parking lot still has enough parking spaces to allocate when faced with sudden, unexpected parking demands. The safety buffer ratio can be calculated based on statistical analysis of prediction deviations in historical data. For example, it can be set according to the standard deviation of historical prediction errors, ensuring the safety buffer covers prediction deviations at a certain confidence level. The system converts available capacity for different time periods. Since the prediction accuracy and uncertainty may differ across time periods, different reduction factors and safety buffer ratios can be configured for different periods. After conversion, the system rounds the result down to ensure the available capacity is an integer.

[0098] A benchmark price is determined based on the hourly rental price of the target parking lot's main business and the prices of similar parking resources in the surrounding market. A price adjustment coefficient is then calculated based on regional demand characteristics, time-of-day vacancy rate characteristics, and price sensitivity characteristics. Finally, a package price range is generated based on the benchmark price and the price adjustment coefficient. The benchmark price serves as the reference basis for package pricing, and its determination method comprehensively considers the parking lot's own pricing level and the price level of the surrounding market. The hourly rental price of the target parking lot's main business reflects its own pricing strategy and customer positioning, while the prices of similar parking resources in the surrounding market reflect the market equilibrium price level in the area. The combination of these two factors ensures that the package pricing is both consistent with the parking lot's own pricing system and competitive in the market. Based on the benchmark price, a price adjustment coefficient is calculated to adjust the benchmark price. The calculation of the price adjustment coefficient is based on the combined effect of regional demand characteristics, time-of-day vacancy rate characteristics, and price sensitivity characteristics.

[0099] The core design logic of the price adjustment coefficient is to enable package prices to dynamically respond to changes in supply and demand and user sensitivity. Higher regional demand indicates a stronger potential demand for parking resources in the surrounding area; in this case, the system increases the price adjustment coefficient, appropriately raising the package price to generate higher revenue. Higher vacancy rates during specific time periods indicate a greater abundance of idle parking spaces and a relative oversupply of resources; in this case, the system decreases the price adjustment coefficient, appropriately lowering the package price to stimulate demand and boost sales. Higher price sensitivity indicates a stronger reaction from the target user group to price adjustments; in this case, the system narrows the price fluctuation range to prevent excessive price changes from causing significant user churn or dissatisfaction. The system generates package price ranges based on the base price and the price adjustment coefficient, with each range including a minimum and a maximum price. The price range setting allows for adjustments during manual confirmation while ensuring its reasonableness through constraints. The system imposes constraints including that the package price is not lower than the cost floor and not higher than the hourly rental price of the main business. The cost floor can be determined based on the marginal cost of parking lot operation to ensure that package sales will not generate losses. The constraint that the package price is not higher than the hourly rental price of the main business ensures that the package is price-attractive to the temporary parking business and will not interfere with the pricing system of the main business.

[0100] The pre-filled parking package draft parameters form a complete package draft that can be manually confirmed later. Package parameters include at least the applicable time period, available capacity, price range, and performance restrictions. The applicable time period is derived from the time period range obtained after classification, aggregation, and cropping operations. The available capacity is derived from integer values ​​converted using reduction factors and safety buffers. The price range is derived from the minimum and maximum prices generated based on the base price and price adjustment factor. Performance restrictions specify various restrictions during package usage, including at least one of the following: entry restrictions, exit restrictions, and applicable license plate range. Entry restrictions specify the time periods during which package users can enter the parking lot; for example, a nighttime package may only allow entry during specific evening hours. Exit restrictions specify the time before which package users must leave the parking lot; for example, a daytime package may require users to leave before a specific evening time. Entry and exit restrictions together constitute the complete time constraints of the package, ensuring that the parking behavior of package users does not exceed the expected resource supply capacity of the parking lot. The applicable license plate range limits which license plate numbers can purchase and use the package. This restriction parameter can be used for targeted marketing or to restrict purchase eligibility, such as limiting purchases to monthly rental users or resident users in specific areas. After all the above processing, a complete draft parking package is generated, including type, priority, applicable time period, available capacity, price range, and various performance restrictions. This draft has met the basic conditions for being put on the market and awaits verification and manual confirmation in subsequent steps.

[0101] Furthermore, you can also view Figure 5 , Figure 5 This is a flowchart illustrating the second embodiment of the parking lot revenue-generating opportunity identification and package generation method of the present invention, based on the shown... Figure 5 Following the step of outputting the parking package draft, steps S70-S80 are also included: Step S70: After the parking package draft is manually confirmed and written into the package pool, obtain the transaction result, performance result and refund result corresponding to the parking package draft; Step S80: Based on the transaction result, the performance result, and the refund result, update the scoring model and the prediction model, wherein the actual parking space occupancy data and the actual available sales data are used as new training samples to calibrate the reduction coefficient of the release capacity and the safety buffer ratio, thereby updating the prediction model.

[0102] This embodiment defines a complete closed-loop feedback optimization mechanism. After the parking package draft is manually confirmed and officially added to the package pool for sale, the mechanism continuously tracks the actual operating performance of the package and feeds back three types of feedback data—transaction results, fulfillment results, and refund results—to the scoring model and prediction model, respectively, driving parameter calibration and performance optimization of the two core models. This closed-loop design ensures that the entire method can continuously evolve based on real operating data, enabling the accuracy of revenue opportunity level determination and idle period prediction to continuously improve over the course of operation.

[0103] Implementation begins after the draft parking package is manually confirmed and added to the package pool. Adding a package to the pool signifies that it has passed all verification and manual review processes and is officially available for purchase by users through the parking lot's sales channels. After the package is listed, its performance is continuously monitored, collecting three types of data from the parking lot's transaction system, access control system, and user management system. The first type is transaction results, which record the complete sales process from listing to sell-out or expiration, including at least one of the following: whether the package was ultimately sold, the percentage of sales volume, and the conversion rate. Whether a package was sold indicates whether an actual purchase occurred. A package might remain unsold due to excessive pricing or insufficient demand, or it might sell out quickly after listing. The percentage of sales volume reflects the proportion of available package capacity that was actually sold. This indicator quantifies the sales progress and market demand intensity. For example, an 80% percentage indicates good sales but with some unsold inventory; a 100% percentage indicates the package is completely sold out; and a very low percentage indicates that market demand for the package is far lower than expected. The conversion rate reflects the percentage of users who see the package and ultimately complete the purchase. This metric comprehensively reflects the degree of match between the package's pricing, time slot settings, and market demand. The second category is fulfillment results, which records various situations that occur during actual user use after the package is sold, including at least one of the following: actual usage rate, fulfillment rate, and time slot conflict rate. The actual usage rate reflects the percentage of sold package shares actually used by users. If users purchase a package but never use it, it indicates that the package settings may not meet their actual needs. The fulfillment rate reflects the percentage of users who complete the package usage according to the agreed rules, such as whether package users enter and leave on time. The time slot conflict rate reflects whether package users exceed the agreed departure time in the parking lot. If a large number of package users do not leave after the agreed departure time, it indicates that the package's time slot settings may be too lenient or the enforcement of the rules may be insufficient. The third category is refund results, which records various situations where users initiate refunds after the package sale, including at least one of the following: refund rate, refund amount percentage, and refund reason distribution. The refund rate reflects the percentage of sold packages that users request refunds. A high refund rate indicates a significant gap between the actual value of the package and user expectations. The refund amount percentage reflects the proportion of refunded amounts to total package sales, while the refund reason distribution records the various reasons users choose when requesting refunds, such as inconvenient timing, excessively high price, or poor user experience. The refund reason distribution is of significant guiding value for diagnosing specific problems with the package plans.

[0104] Based on the collected transaction results, fulfillment results, and refund results, the scoring model and the prediction model are updated respectively. The scoring model is the core model used to determine the level of revenue-generating opportunities, while the prediction model is the core model used to predict the distribution of idle periods. The update methods of the two models are significantly different. This differentiated design stems from the different roles that the two types of models play in the overall methodology and the different data foundations they rely on.

[0105] For updating the scoring model, the actual conversion rate or actual revenue increase is used as a supervisory label to correct the model. The essence of the scoring model is to rank and classify the revenue potential of each parking lot based on multi-dimensional features, while the actual conversion rate and actual revenue increase are the most direct indicators of whether the revenue potential has been successfully realized. If a parking lot is judged to have a high opportunity level by the scoring model, but its actual conversion rate after the packages are listed is extremely low or the revenue increase is far below expectations, it indicates a deviation in the weighting of each dimension in the scoring model or in the mapping relationship between the overall score and the opportunity level. In this case, the multi-dimensional feature data of the parking lot and its actual conversion rate are used to form a labeled training sample to refit the parameters of the scoring model. Specifically, a regression model can be retrained using linear regression or gradient boosting trees, with the normalized scores of each dimension as independent variables and the actual conversion rate as the dependent variable, replacing the original weights of each dimension with the newly trained regression coefficients. The system can also adjust the threshold division between the comprehensive score and the opportunity level by comparing the distribution differences in actual transaction conversion rates between high-opportunity level samples and low-opportunity level samples, making the consistency between the scoring model's level determination results and actual transaction performance higher. This parameter correction method based on real transaction results enables the scoring model to continuously learn the actual revenue increase effect corresponding to different feature combinations, thereby providing more accurate level determinations in subsequent parking lot revenue increase opportunity identification.

[0106] For updating the prediction model, actual parking space occupancy data and actual available parking space data are used as new training samples to calibrate the reduction factor and safety buffer ratio of the release capacity. The idle time distribution and release capacity output by the prediction model are essentially forward-looking estimates of future parking space occupancy; inevitably, there will be a certain degree of deviation between these estimates and the actual parking space occupancy. The system adds actual parking space occupancy data collected during the actual sale and use of packages as new training samples to the prediction model's training set, using actual available parking space data as the validation standard to re-evaluate the deviation distribution characteristics of the prediction model. Specifically, the release capacity predicted by the prediction model for a certain period is compared with the actual release capacity, and the deviation rate between the two is calculated. If the deviation rate shows a systematically high trend, it indicates that the prediction model tends to overestimate the sufficiency of idle resources. In this case, the system correspondingly reduces the value of the reduction factor, making the conversion result of the available parking capacity more conservative. Conversely, if the deviation rate shows a systematically low trend, it indicates that the prediction model tends to underestimate the sufficiency of idle resources. In this case, the reduction factor should be increased accordingly to make the conversion result of available capacity more positive. This parameter can also be calibrated based on whether the safety buffer ratio is frequently triggered. If the number of parking spaces actually occupied after the package is sold often exceeds the predicted value of the release capacity, resulting in a shortage of parking spaces, it indicates that the safety buffer ratio is set too low, and the value of the safety buffer ratio should be increased accordingly. Through continuous calibration with actual data, the prediction deviation of the prediction model can converge quickly in several iterations, making its predictions of the distribution of idle time periods and release capacity increasingly closer to the actual operating conditions.

[0107] The scoring and prediction models can be updated using a combination of periodic retraining and triggered updates. Periodic retraining involves retraining the model fully or incrementally with the latest window data at preset intervals. Triggered updates are triggered immediately when the prediction error or transaction deviation exceeds a preset threshold. During the update process, learning rate decay and regularization are used to prevent overfitting and catastrophic forgetting. Weight updates are achieved through gradient descent or Bayesian posterior updates, and upper limits are set on the magnitude of each update or smoothing is performed to ensure model stability. Through a closed-loop feedback mechanism, the entire method achieves a closed-loop process from data collection, feature extraction, opportunity identification, package generation to effect feedback and model optimization, forming a continuous self-evolution capability.

[0108] Furthermore, you can also view Figure 6 , Figure 6 This is a flowchart illustrating the third embodiment of the parking lot revenue-generating opportunity identification and package generation method of the present invention, based on the shown... Figure 6 Following the step of acquiring multi-source operating data of the target parking lot, steps S90~S130 are further included: Step S90: Perform anomaly identification on the multi-source operating data and remove abnormal data records that do not meet the preset threshold range; Step S100: Perform time-series alignment on the operational data from each source after removing abnormal data at a uniform time granularity, and interpolate and fill in the missing time slices in the aligned data. Step S110: Calculate the data integrity rate of each preset time granularity based on the data after interpolation and filling. When the data integrity rate is lower than the preset integrity rate threshold, mark the operating data under the corresponding time granularity as unavailable and trigger a data supplementation command. Step S120: Record the time granularity information marked as unavailable to the data quality log. The data quality log is used to skip data of the corresponding time granularity during subsequent feature extraction. Step S130: Store the multi-source business data processed as described above into the cleaned business database.

[0109] This embodiment further refines the data preprocessing workflow after the multi-source operational data acquisition stage and before the feature extraction stage. This workflow performs a series of cleaning operations on the collected raw multi-source operational data, including anomaly identification and removal, temporal alignment and missing data imputation, data integrity verification, and quality labeling. Finally, the verified reliable data is stored in the cleaned operational database for subsequent steps. The core objective of this series of operations is to ensure the data quality input to the feature extraction stage, avoiding negative impacts on the accuracy of subsequent opportunity identification and prediction due to outliers, missing values, or temporal inconsistencies in the raw data.

[0110] In this embodiment, anomaly identification is performed on the collected multi-source operational data, and abnormal data records that do not conform to the preset threshold range are removed. The necessity of anomaly identification lies in the fact that the collection process of parking lot operational data involves multiple data sources and various collection devices. These devices and systems may generate abnormal data deviating from the normal range due to equipment failure, network fluctuations, manual input errors, etc., during actual operation. A preset threshold range is set for each type of operational data, determined based on the distribution characteristics of historical normal data. For the historical entry and exit timestamp data, a reasonable range for the total daily entry volume and a reasonable fluctuation range for the hourly entry volume are set. When the entry volume in a certain period far exceeds or falls far below the historical normal level for that period, the system identifies the record as abnormal data. For parking space occupancy data, the system sets a reasonable range for the occupancy rate between 0% and 100%. If the parking space occupancy rate at a certain time granularity exceeds this physical boundary, the data is obviously abnormal. For hourly rental price records in the charging data, the system sets a reasonable price range based on the parking lot's historical pricing level and similar market prices. Price records exceeding this range are identified as abnormal. For historical transaction data, reasonable ranges for transaction quantity and amount are set based on the historical transaction volume of the parking lot. Regional demand data is based on thresholds set according to the normal demand level of the target parking lot's area. After anomaly identification, abnormal data records that do not meet the threshold ranges are removed from the dataset. This removal operation only removes abnormal data points without affecting normal data from other sources at the same time granularity. For the removed abnormal data, the system records its original value, time granularity, and reason for the anomaly. These records can be used for subsequent data quality traceability and root cause analysis.

[0111] After removing outlier data, operational data from various sources are time-series aligned to a unified time granularity, and missing time slices in the aligned data are imputed. Different data sources may have different record granularities. For example, entry and exit data might be real-time event data for each transaction, parking space occupancy data might be snapshot data collected every five minutes, fee and transaction data might be daily aggregated statistics, and regional demand data might be external data updated hourly. The time-series alignment process converts these raw data with different granularities and time bases into a unified time granularity. First, a standard time granularity is determined as the alignment benchmark. This standard time granularity can be set to hourly or half-hourly based on the needs of subsequent feature extraction and predictive modeling. Then, the system resamples and aggregates the raw data from each source according to the standard time granularity, mapping the event time or record time in the raw data to its corresponding time slice, and performing necessary statistical aggregation on data points within the same time slice.

[0112] After time-series alignment is completed, the aligned data sequence is checked for missing time slices. Possible reasons for missing time slices include the original data not being recorded within that time slice, an interruption occurring during data acquisition, or the data in that time slice being removed as outlier data. For missing time slices, imputation is performed. The imputation method can be flexibly selected based on the type of missing data and data characteristics. When a missing time slice occurs in isolation and the data in the preceding and following time slices are normal, linear imputation is used, i.e., the average of the data in the preceding and following time slices is used as the imputation value for that time slice. When multiple missing time slices occur consecutively, the historical average of the same time slice on the same date type is used for imputation. For example, if the missing time slice is 10:00 AM on a weekday, the average parking occupancy rate at 10:00 AM on all weekdays over the past few weeks is used as the imputation value. When a missing time slice belongs to a specific holiday or special date, the historical trend of the same period under that date type is used for imputation. For missing time slices that cannot be effectively imputed using the above methods, they are retained as missing for further processing.

[0113] The data integrity rate for each preset time granularity is calculated based on the imputed data. When the data integrity rate is lower than the preset integrity rate threshold, the operational data for the corresponding time granularity is marked as unavailable, and a data replenishment command is triggered. The data integrity rate is evaluated for all data sources for each time granularity, reflecting the proportion of valid data in the operational data from each source at that time granularity. The system iterates through each time granularity, counts the number of data items from each source that are not missing and have not been anomaly removed, and divides this number by the total number of complete data items that the time granularity should have to obtain the data integrity rate. The data integrity rate ranges from zero to 100%, with a higher integrity rate indicating more complete and reliable data for that time granularity. A preset integrity rate threshold is set; when the data integrity rate for a certain time granularity is lower than this threshold, the data for that time granularity is considered insufficient to support subsequent feature extraction and modeling analysis. For time granularities marked as unavailable, a data replenishment command is triggered, which notifies the data acquisition module to attempt to retrieve the data for that time granularity from the original data source or sends a data replenishment reminder to operations personnel. Before the data replenishment is completed, the operational data at this time granularity remains unavailable and will not participate in subsequent feature extraction steps.

[0114] The time granularity information of data marked as unavailable is recorded in the data quality log. The data quality log is a continuously accumulating record of data quality, containing at least the identifier of the marked time granularity, the data integrity rate value for that time granularity, the preset integrity rate threshold, the marking time, and the duration of the unavailable state. The core value of the data quality log lies in providing clear jump instructions for subsequent feature extraction steps. When performing feature extraction, the data quality log is queried first. When calculating statistical features across different time granularities, time granularities marked as unavailable in the log are skipped. Skipping this step means that the data at that time granularity is not involved in any mean calculation, variance calculation, extreme value calculation, or periodic statistical analysis. This mechanism ensures that low-quality data marked as unavailable will not mislead subsequent opportunity identification and prediction due to its participation in feature extraction. The data quality log also allows operations personnel to review it regularly, helping them understand the data quality status of each parking lot at different times, identify weaknesses in the data collection process, and take targeted improvement measures.

[0115] Multi-source operational data, after undergoing a series of preprocessing operations including anomaly identification and removal, temporal alignment and imputation, data integrity verification and quality marking, and quality log recording, is stored in a cleaned operational database. The operational database is a structured data collection where data is organized according to a uniform time granularity. The operational data from each source at each time granularity undergoes quality marking to verify its reliability. Data stored in the operational database falls into two categories: data marked as usable, which will be used normally in subsequent feature extraction steps; and data marked as unusable, which, although also stored in the database, will be skipped during feature extraction. The operational database also stores the original source information and processing logs of each data point, supporting subsequent data traceability and auditing needs. Through this complete cleaning process, the data in the operational database possesses good temporal consistency, integrity, and reliability, providing a high-quality data foundation for multi-dimensional feature extraction.

[0116] Furthermore, you can also view Figure 7 , Figure 7 A schematic diagram of the system structure, such as Figure 1 The diagram illustrates the modular composition of the entire system and the data interaction relationships between the modules. The core modules of the system include a data acquisition module, a feature extraction module, an opportunity identification module, a time period prediction module, a package draft generation module, and a confirmation flow module. These modules are connected in series to form a complete processing chain from raw data input to package draft output.

[0117] The data acquisition module, serving as the system's data entry point, establishes data connections with the target parking lot's parking management system, parking space detection equipment, charging system, and external data sources. It acquires entry and exit time-series data, parking space occupancy data, charging data, historical transaction data, and regional demand data, storing the collected raw data in the operational database. The feature extraction module reads cleaned multi-source operational data from the operational database, organizes the data according to a unified time granularity, and extracts features such as time-period vacancy rate, time-period repetition, regional demand intensity, price sensitivity, and revenue fluctuation. These features collectively constitute a multi-dimensional feature vector reflecting the parking lot's revenue growth potential.

[0118] The multi-dimensional features output by the feature extraction module flow to the opportunity identification module and the time period prediction module, respectively. After receiving the multi-dimensional features, the opportunity identification module performs normalization processing on each dimension, calculates a comprehensive score through weighted summation, and classifies the target parking lot into high-opportunity, medium-opportunity, or low-opportunity levels based on a preset threshold, determining the corresponding parking priority. The time period prediction module receives the time-series related features from the multi-dimensional features, combines them with historical parking space occupancy rate sequences and calendar features, and outputs the distribution of idle time periods and the corresponding release capacity for each time period within a preset future period through a time-series prediction model.

[0119] The revenue-increasing opportunity level output by the opportunity identification module and the idle time distribution output by the time period prediction module, along with the available release capacity, are input into the package draft generation module. The package draft generation module determines the package type and priority based on the revenue-increasing opportunity level, categorizes the idle time distribution into weekday daytime hours, weekday nighttime hours, weekend hours, or holiday hours, calculates the package's saleable capacity based on the available release capacity, generates a price range based on the benchmark price and supply-demand adjustment coefficient, and pre-fills in performance restriction parameters such as entry restrictions, exit restrictions, and applicable license plate ranges to form a complete parking package draft.

[0120] The draft package generated by the module first undergoes a verification process, which includes at least time slot conflict verification and capacity constraint verification. Once verified, the draft package is sent to the confirmation and workflow module. This module presents the draft package in a structured confirmation interface to authorized personnel. After confirmation, modification, or rejection, the approved draft package is written to the package pool. After being written to the pool, the confirmation and workflow module continuously tracks the package's transaction results, fulfillment results, and refund results, and sends feedback data back to the opportunity identification module and the time slot prediction module. This drives parameter calibration and performance optimization of the two core models, forming a complete closed loop from data collection, feature extraction, opportunity identification, package generation to effect feedback.

[0121] Furthermore, you can also view Figure 8 , Figure 8 This is a diagram illustrating the process of generating a draft package and obtaining manual confirmation. Figure 3 The document details the logic behind the package proposal generation and manual confirmation process. This process takes the revenue-increasing opportunity level output by the opportunity identification module and the idle time distribution and releaseable capacity output by the time period prediction module as input. It goes through three stages: automatic package proposal generation, multi-dimensional verification, and manual confirmation, ultimately outputting the confirmed package proposal and writing it into the package pool. This process is the core of the entire methodology, transforming data analysis results into executable business solutions.

[0122] The automatic generation of package proposals first receives the revenue-generating opportunity level. Based on the level, the package type and priority are determined. High opportunity levels are matched with high-value, long-term package types and high listing priority; medium opportunity levels are matched with regular package types and medium listing priority; and low opportunity levels are matched with conservative, short-term package types and low listing priority or delayed listing. Simultaneously, the system categorizes idle time slots according to preset time slot types, mapping them to at least one of weekday daytime, weekday nighttime, weekend, or holiday time slots. Consecutive or adjacent idle time slots are aggregated into applicable time slot intervals for the packages, removing time slots overlapping with peak business hours during the aggregation process. The conversion of saleable capacity is done by multiplying the releaseable capacity by a reduction factor, deducting a safety buffer, and rounding down. The reduction factor is related to the revenue-generating opportunity level; higher opportunity levels use higher reduction factors to fully release resources, while lower opportunity levels use lower reduction factors to control risk. The price range is generated based on a base price determined by the main business hourly rental price and surrounding market prices. A price adjustment coefficient is then calculated based on regional demand, vacancy rate during specific time periods, and price sensitivity. A higher coefficient increases demand, causing the price to rise; a higher vacancy rate decreases the coefficient, causing the price to fall; and higher price sensitivity narrows the fluctuation range. Simultaneously, constraints are imposed on the price range to ensure that the package's converted unit price is not lower than the cost floor and not higher than the main business hourly rental price. The system then pre-fills in the applicable time period, available capacity, price range, and fulfillment restriction parameters. Fulfillment restriction parameters include at least one of entry restrictions, exit restrictions, and applicable license plate ranges, forming a complete package draft.

[0123] After the package draft is automatically generated, it enters a multi-dimensional verification process. The system first performs a time-slot conflict check, checking if the applicable time slots of the package overlap with peak periods of the main business. If there is an overlap, the time slots are adjusted or conflict information is marked. Then, a capacity constraint check is performed, checking if the package's saleable capacity exceeds the releaseable capacity for the corresponding time slot. If it does, it is automatically reduced to within the releaseable capacity range. Simultaneously, it checks if the cumulative saleable capacity of multiple packages within the same time slot exceeds the total releaseable capacity for that time slot. If it does, capacity allocation or adjustment is performed according to the priority of each package. In some implementations, main business price impact checks and performance risk checks can also be performed. Any draft that fails any check records the specific reason for failure and attempts automatic correction according to preset correction rules. If it still fails after automatic correction, it is marked as awaiting manual processing.

[0124] After verification, the confirmation workflow module organizes the draft package into a structured confirmation interface and outputs it to the operator. The interface fully displays the entire content and key parameters of the draft package, along with information on the basis for its generation, such as the revenue increase opportunity level, comprehensive score, detailed normalized scores for each dimension, idle period prediction confidence level, and verification results. Editable fields are also provided for operators to adjust parameters. Operators can review the package through three options: confirm, modify, or reject. When confirming, the system writes the draft to the package pool and pushes it to the sales channel for listing. When modifying, the system receives the changes and re-executes verification; if successful, it is written to the package pool. When rejecting, the system records the reason for rejection and archives the draft in the history. All operation records are fully logged in the system log. After the package is written to the package pool, the confirmation workflow module continuously tracks the transaction results, fulfillment results, and refund results of the package, providing a data foundation for subsequent model feedback updates and forming a complete business loop from generation to confirmation to effect feedback.

[0125] This application provides a parking lot revenue-generating opportunity identification and package generation device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the parking lot revenue-generating opportunity identification and package generation method in the above embodiment 1.

[0126] The following is for reference. Figure 9 The diagram illustrates a structural schematic of a parking lot revenue-generating opportunity identification and package generation device suitable for implementing embodiments of this application. The parking lot revenue-generating opportunity identification and package generation device in embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 9 The parking lot revenue-generating opportunity identification and package generation device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0127] like Figure 9As shown, the parking fee increase opportunity identification and package generation device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the parking fee increase opportunity identification and package generation device. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the parking fee opportunity identification and package generation device to exchange data wirelessly or via wired communication with other devices. Although parking fee opportunity identification and package generation devices with various systems are shown in the figures, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems may be implemented alternatively.

[0128] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0129] The parking lot revenue-generating opportunity identification and package generation device provided in this application, employing the parking lot revenue-generating opportunity identification and package generation method described in the above embodiments, can solve the technical problem that the identification of revenue-generating opportunities and package generation in the existing parking resource revenue-generating management process highly relies on manual judgment and lacks an automated closed-loop processing mechanism. Compared with the prior art, the beneficial effects of the parking lot revenue-generating opportunity identification and package generation device provided in this application are the same as those of the parking lot revenue-generating opportunity identification and package generation method provided in the above embodiments, and other technical features in this parking lot revenue-generating opportunity identification and package generation device are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0130] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0131] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0132] This application provides a storage medium, which is a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, which are used to execute the parking lot revenue increase opportunity identification and package generation method in the above embodiments.

[0133] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0134] The aforementioned computer-readable storage medium may be included in the parking lot revenue-generating opportunity identification and package generation device; or it may exist independently and not be assembled into the parking lot revenue-generating opportunity identification and package generation device.

[0135] The aforementioned computer-readable storage medium carries one or more programs. When the aforementioned one or more programs are executed by the parking lot revenue-generating opportunity identification and package generation device, the parking lot revenue-generating opportunity identification and package generation device implements the technical content of the parking lot revenue-generating opportunity identification and package generation method embodiment shown above.

[0136] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof. These programming languages ​​include object-oriented programming languages—such as Java, Smalltalk, and C++—and conventional procedural programming languages—such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0137] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0138] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0139] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described parking lot revenue-generating opportunity identification and package generation method. This solves the technical problem that existing parking resource revenue-generating management processes heavily rely on manual judgment for revenue-generating opportunity identification and package generation, lacking an automated closed-loop processing mechanism. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the parking lot revenue-generating opportunity identification and package generation method provided in the above embodiments, and will not be repeated here.

Claims

1. A method for identifying parking lot revenue-generating opportunities and generating service packages, characterized in that, Includes the following steps: Acquire multi-source operational data of the target parking lot, and extract multi-dimensional features from the multi-source operational data; The revenue-generating opportunity level of the target parking lot is determined based on the aforementioned multi-dimensional features; Based on the time-related features in the multi-dimensional features, predict the distribution of idle time periods and releaseable capacity within a future preset period; Based on the revenue-increasing opportunity level, the idle time distribution, and the available capacity, a draft parking package is generated and package parameters are pre-filled. The proposed parking package is validated, and the validation includes at least time period conflict validation and capacity constraint validation. Output the draft parking package after verification.

2. The parking lot revenue-generating opportunity identification and package generation method as described in claim 1, characterized in that, The step of determining the revenue-generating opportunity level of the target parking lot based on the multi-dimensional features includes: Normalization is performed on each dimension of the multi-dimensional features to obtain the normalized score of each dimension. The multi-dimensional features include time period vacancy rate feature, time period repetition feature, regional demand heat feature, price sensitivity feature and income fluctuation feature. Each dimension is assigned a weight, which is determined by at least one of the following methods: expert weighting, entropy weighting, or data-driven weighting. The comprehensive score is calculated by weighting and summing the normalized scores of each dimension with their corresponding weights. Based on the comprehensive score and the preset classification threshold, the target parking lot is divided into multiple revenue-increasing opportunity levels, including high opportunity, medium opportunity and low opportunity. The corresponding listing priority is determined based on the revenue increase opportunity level obtained from the classification. The high opportunity corresponds to the high listing priority, the medium opportunity corresponds to the medium listing priority, and the low opportunity corresponds to the low listing priority or postponement of listing.

3. The parking lot revenue-generating opportunity identification and package generation method as described in claim 2, characterized in that, After the step of dividing the target parking lot into multiple revenue-generating opportunity levels based on the comprehensive score and a preset classification threshold, the method further includes: Obtain the normalized score of at least one key dimension of the target parking lot; When the normalized score of the key dimension is lower than the preset lower limit, the income increase opportunity level will be downgraded by one or more levels. The key dimension includes at least the regional demand heat feature.

4. The parking lot revenue-generating opportunity identification and package generation method as described in claim 1, characterized in that, The step of predicting the distribution of idle time periods and releaseable capacity within a future preset period based on the time-series correlation features in the multi-dimensional features includes: Obtain historical parking space occupancy rate time series data of the target parking lot; Based on the historical parking space occupancy rate time series data and calendar features, the idle probability or number of idle parking spaces at each time granularity within a future preset period is predicted using a time series prediction model. Consecutive or adjacent idle time slices are aggregated into idle time intervals, each of which includes a start and end time. The corresponding release capacity is determined based on the number of idle parking spaces within each idle time period interval, and the calendar features include at least one of time, day of the week, weekday identifier, weekend identifier, and holiday identifier.

5. The parking lot revenue-generating opportunity identification and package generation method as described in claim 4, characterized in that, The step of predicting the probability of vacancy or the number of vacant parking spaces at each time granularity within a future preset period based on the historical parking space occupancy rate time series data and calendar features using a time series prediction model includes: Obtain regional demand data, surrounding points of interest data, and vacancy data of competitor parking lots for the target parking lot; The regional demand data, the surrounding points of interest data, and the vacancy data of competitor parking lots are used as exogenous variables and input into the time series prediction model. The time series prediction model is any one of a time series statistical model, a gradient boosting model, or a recurrent neural network. When the parking space occupancy data of the target parking lot shows periodicity and stable regularity, the time series statistical model is selected. When the input feature dimension contains exogenous variables and there is a non-linear interaction relationship between features, the gradient boosting model is selected. When the time series has a long-range dependency relationship and the sample size is sufficient, the recurrent neural network is selected.

6. The parking lot revenue-generating opportunity identification and package generation method as described in claim 1, characterized in that, The step of generating a draft parking package and pre-filling package parameters based on the revenue increase opportunity level, the idle time distribution, and the available capacity includes: The package type and package priority are determined based on the aforementioned revenue-increasing opportunity level; The idle time periods are categorized according to preset time period types and mapped to at least one of weekday daytime periods, weekday nighttime periods, weekend periods, or holiday periods. Continuous or adjacent idle time periods are aggregated into time periods applicable to the package. The saleable capacity of the package is calculated by converting the releasable capacity into the package capacity. The saleable capacity of the package is obtained by multiplying the releasable capacity by a reduction factor and deducting the safety buffer, and then rounding it to the nearest whole number. A benchmark price is determined based on the hourly rental price of the target parking lot's main business and the prices of similar parking resources in the surrounding market. A price adjustment coefficient is calculated based on the regional demand characteristics, the time period vacancy rate characteristics, and the price sensitivity characteristics. A package price range is generated based on the benchmark price and the price adjustment coefficient. The applicable time period, available capacity, price range, and performance restriction parameters of the parking package draft are pre-filled. The performance restriction parameters include at least one of the following: entry restrictions, exit restrictions, and applicable license plate range.

7. The parking lot revenue-generating opportunity identification and package generation method as described in claim 1, characterized in that, Following the step of verifying the parking package draft, the following steps are also included: After the parking package draft is manually confirmed and written into the package pool, the transaction result, performance result and refund result corresponding to the parking package draft are obtained; Based on the transaction results, performance results, and refund results, the scoring model and prediction model are updated. Specifically, the actual parking space occupancy data and actual available sales data are used as new training samples to calibrate the reduction coefficient of releaseable capacity and the safety buffer ratio, thereby updating the prediction model.

8. The parking lot revenue-generating opportunity identification and package generation method as described in claim 1, characterized in that, Following the step of acquiring multi-source operating data of the target parking lot, the method further includes: The multi-source operational data is anomaly identified, and abnormal data records that do not meet the preset threshold range are removed; After removing outlier data, the operational data from each source are aligned in time series at a uniform time granularity, and missing time slices in the aligned data are interpolated and filled. The data integrity rate of each preset time granularity is calculated based on the data after interpolation and filling. When the data integrity rate is lower than the preset integrity rate threshold, the operating data under the corresponding time granularity is marked as unavailable and a data supplementation command is triggered. The time granularity information marked as unavailable is recorded in the data quality log, which is used to skip data of the corresponding time granularity during subsequent feature extraction; The multi-source business data processed as described above is stored in the cleaned business database.

9. A parking lot revenue-generating opportunity identification and package generation device, characterized in that, The parking lot revenue-generating opportunity identification and package generation device stores a computer program, which, when executed by a processor, implements the parking lot revenue-generating opportunity identification and package generation method according to any one of claims 1-8.

10. A storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the parking lot revenue-generating opportunity identification and package generation method according to any one of claims 1-8.

Images